Saturday, March 06, 2010

The Forecasting Problem

The forecasting problem is integral to successful enterprise and public administration. Yet, many senior executives, managers, and administrators have difficulty interpreting the meaning of a given time-series forecast based on methodology alone. For example, consider the sequence of forecasts that follow; each 30-period extended forecast was created in Excel (Microsoft) using identical historical data with varying methods.

The first chart above depicts a straight linear forecast through the historical data and into the future. In this instance, the trend line slopes downward.

The second chart depicts a logarithmic forecast. Note that the 30-period future forecast here is slightly higher than in the previous example.

The next chart above illustrates the use of a third order polynomial and projects a significantly higher 30-period outcome than the previous methods.

This fourth chart depicts a fourth order polynomial trend line predicting declining results in the future.

The last graph above is a fifth order polynomial forecast and predicts that results will decline sharply in the future.

Observe that the forecasts generated by each method are quite different from each other. Moreover, the methods depicted above illustrate only a small sampling of the most common time-series forecasting techniques in use today. Other forecasting methods include such techniques as autoregressive moving averages (ARMA), generalized autoregressive conditional heteroskedastic methods (GARCH), multivariate forecasting methods, and a long list of other advanced techniques in frequent use by companies and governments around the world.

Given the above, the challenge for executives and managers becomes how best to compare and evaluate the various forecasting methods in use today. A number of research questions come to mind: What criteria does one employ to differentiate and evaluate a specific forecast? What specific conceptual methodologies (e.g., mathematical) underlie each approach? Finally, what message does one send (or not) in choosing a specific forecasting technique (and projection)?

The forecasting problem in enterprise and government is very real, and simply training more analysts to forecast effectively is only part of the solution. We also need more senior executives, managers, journalists, public administrators (including politicians), and leaders in general who can comprehend and interpret the strengths, weakness, and implications of the various state-of-the-art forecasting techniques in use, today. Finally, the public at large would benefit from increased comprehension and understanding of modern forecasting methods as a listening audience, if only to avert poor judgments in their purchasing, advocacy, and voting decisions.

I am reminded of a story told to me by a professor many years ago about the importance of understanding statistical forecasting. The story begins in a US Civil War setting back in 1864 with an officer reading and paraphrasing headlines from a newspaper to some soldiers who had gathered around eager to hear the news of the day.

"It says here that we suffered 50 percent casualties in Vicksburg the other day...," said the officer.

Upon hearing the report, a nearby soldier responded to the officer by asking, "Wow, is that a lot...?"

The moral of the story is that statistical forecasting is everyone's business, but most especially for the management of enterprise in the 21st century.

To learn more about other more advanced time-series forecasting methods, including training and software tools, visit my website linked below.

Learn More

Related Articles:

Tools for Decision and Risk Analysis


Rug said...


Really interesting post.

I agree 100% that decision makers have to understand the limits and especially the possibilities that forecasting offers.

In my opinion that means three things:

First, the will to seek out the truth; too many managers want tools to confirm what they think is true and not tools to help them understand what may be true.

Second an understanding of where data comes from and what it represents is fundamental to good forecasting (never a simple task).

Third, a good handle on forecasting techniques. Creating and using appropriate models is an admixture of art and science, and using multiple techniques is usually a good approach (this article by J. Scott Armstrong is quite interesting assuming that you understand the underlying implications of different models and methods.



Joe Rotger said...

Overadjusting is a killer --a straight line is a lot better.
Downtrending was broken, stayed within a channel with a failed attempt to break upwards...

Post a Comment